Table 13 Correctly Predicted-Accepted (CPA) and Incorrectly Predicted-Rejected (IPR) under FGSM and PGD adversarial attacks across uncertainty measures.

From: Detection of AI generated images using combined uncertainty measures and particle swarm optimised rejection mechanism

Dataset	Uncertainty Measure	CPA	IPR
FGSM \(\epsilon =0.01\)	Total Fisher Information	97.30	2.63
	Fisher Frobenius Norm	97.32	2.77
	Fisher Entropy	97.62	9.59
	Gaussian Process	94.80	50.84
	Entropy of Expected	100.00	46.86
	Knowledge Uncertainty	100.00	41.98
	Combined Uncertainty	91.70	42.94
FGSM \(\epsilon =0.03\)	Total Fisher Information	95.02	3.73
	Fisher Frobenius Norm	95.11	3.74
	Fisher Entropy	95.54	9.00
	Gaussian Process	92.92	67.07
	Entropy of Expected	100.00	46.15
	Knowledge Uncertainty	100.00	46.01
	Combined Uncertainty	91.84	52.97
FGSM \(\epsilon =0.05\)	Total Fisher Information	93.81	3.33
	Fisher Frobenius Norm	93.96	3.31
	Fisher Entropy	95.48	4.95
	Gaussian Process	91.34	65.08
	Entropy of Expected	100.00	48.99
	Knowledge Uncertainty	100.00	51.29
	Combined Uncertainty	91.78	50.68
PGD \(\epsilon =0.01\)	Total Fisher Information	0.00	100.00
	Fisher Frobenius Norm	0.00	100.00
	Fisher Entropy	100.00	0.02
	Gaussian Process	69.96	99.73
	Entropy of Expected	100.00	27.79
	Knowledge Uncertainty	100.00	24.50
	Combined Uncertainty	93.74	74.72
PGD \(\epsilon =0.03\)	Total Fisher Information	0.00	100.00
	Fisher Frobenius Norm	0.00	100.00
	Fisher Entropy	0.00	100.00
	Gaussian Process	78.21	100.00
	Entropy of Expected	100.00	28.31
	Knowledge Uncertainty	100.00	24.75
	Combined Uncertainty	92.74	73.10
PGD \(\epsilon =0.05\)	Total Fisher Information	0.00	100.00
	Fisher Frobenius Norm	0.00	100.00
	Fisher Entropy	99.98	0.00
	Gaussian Process	77.97	100.00
	Entropy of Expected	100.00	28.46
	Knowledge Uncertainty	100.00	25.05
	Combined Uncertainty	91.23	74.03

Bold values indicate the highest Correctly Predicted-Accepted (CPA) and highest Incorrectly Predicted-Rejected (IPR) rates.

Back to article page

Quick links

Search

Quick links